HAL interface for compilation and execution hints
The following AIDL types are added:
- TokenValuePair
- PrepareModelConfig
- ExecutionConfig
The following AIDL methods are added:
- IDevice::prepareModelWithConfig
- IPreparedModel::executeSynchronouslyWithConfig
- IPreparedModel::executeFencedWithConfig
- IBurst::executeSynchronouslyWithConfig
The compilation and execution hints are being stored as a list of
token-value pairs as part of the PrepareModelConfig / ExecutionConfig.
And the PrepareModelConfig / ExecutionConfig parcelables are created in
order to make future extensions to the execution related interfaces
easier.
It is the drivers responsibility to verify the hints, and it is allowed
for the driver to ignore them.
Bug: 203248587
Test: neuralnetworks_utils_hal_aidl_test
Change-Id: I98240fd75089fc85cdfcaa0be28aab8a6f0dfca5
diff --git a/neuralnetworks/utils/common/src/ResilientPreparedModel.cpp b/neuralnetworks/utils/common/src/ResilientPreparedModel.cpp
index 1ae19bc..b5843c0 100644
--- a/neuralnetworks/utils/common/src/ResilientPreparedModel.cpp
+++ b/neuralnetworks/utils/common/src/ResilientPreparedModel.cpp
@@ -104,43 +104,53 @@
}
nn::ExecutionResult<std::pair<std::vector<nn::OutputShape>, nn::Timing>>
-ResilientPreparedModel::execute(const nn::Request& request, nn::MeasureTiming measure,
- const nn::OptionalTimePoint& deadline,
- const nn::OptionalDuration& loopTimeoutDuration) const {
- const auto fn = [&request, measure, &deadline,
- &loopTimeoutDuration](const nn::IPreparedModel& preparedModel) {
- return preparedModel.execute(request, measure, deadline, loopTimeoutDuration);
+ResilientPreparedModel::execute(
+ const nn::Request& request, nn::MeasureTiming measure,
+ const nn::OptionalTimePoint& deadline, const nn::OptionalDuration& loopTimeoutDuration,
+ const std::vector<nn::TokenValuePair>& hints,
+ const std::vector<nn::ExtensionNameAndPrefix>& extensionNameToPrefix) const {
+ const auto fn = [&request, measure, &deadline, &loopTimeoutDuration, &hints,
+ &extensionNameToPrefix](const nn::IPreparedModel& preparedModel) {
+ return preparedModel.execute(request, measure, deadline, loopTimeoutDuration, hints,
+ extensionNameToPrefix);
};
return protect(*this, fn);
}
nn::GeneralResult<std::pair<nn::SyncFence, nn::ExecuteFencedInfoCallback>>
-ResilientPreparedModel::executeFenced(const nn::Request& request,
- const std::vector<nn::SyncFence>& waitFor,
- nn::MeasureTiming measure,
- const nn::OptionalTimePoint& deadline,
- const nn::OptionalDuration& loopTimeoutDuration,
- const nn::OptionalDuration& timeoutDurationAfterFence) const {
+ResilientPreparedModel::executeFenced(
+ const nn::Request& request, const std::vector<nn::SyncFence>& waitFor,
+ nn::MeasureTiming measure, const nn::OptionalTimePoint& deadline,
+ const nn::OptionalDuration& loopTimeoutDuration,
+ const nn::OptionalDuration& timeoutDurationAfterFence,
+ const std::vector<nn::TokenValuePair>& hints,
+ const std::vector<nn::ExtensionNameAndPrefix>& extensionNameToPrefix) const {
const auto fn = [&request, &waitFor, measure, &deadline, &loopTimeoutDuration,
- &timeoutDurationAfterFence](const nn::IPreparedModel& preparedModel) {
+ &timeoutDurationAfterFence, &hints,
+ &extensionNameToPrefix](const nn::IPreparedModel& preparedModel) {
return preparedModel.executeFenced(request, waitFor, measure, deadline, loopTimeoutDuration,
- timeoutDurationAfterFence);
+ timeoutDurationAfterFence, hints, extensionNameToPrefix);
};
return protect(*this, fn);
}
nn::GeneralResult<nn::SharedExecution> ResilientPreparedModel::createReusableExecution(
const nn::Request& request, nn::MeasureTiming measure,
- const nn::OptionalDuration& loopTimeoutDuration) const {
+ const nn::OptionalDuration& loopTimeoutDuration,
+ const std::vector<nn::TokenValuePair>& hints,
+ const std::vector<nn::ExtensionNameAndPrefix>& extensionNameToPrefix) const {
#if 0
auto self = shared_from_this();
- ResilientExecution::Factory makeExecution =
- [preparedModel = std::move(self), request, measure, loopTimeoutDuration] {
- return preparedModel->createReusableExecutionInternal(request, measure, loopTimeoutDuration);
+ ResilientExecution::Factory makeExecution = [preparedModel = std::move(self), request, measure,
+ loopTimeoutDuration, hints,
+ extensionNameToPrefix] {
+ return preparedModel->createReusableExecutionInternal(request, measure, loopTimeoutDuration,
+ hints, extensionNameToPrefix);
};
return ResilientExecution::create(std::move(makeExecution));
#else
- return createReusableExecutionInternal(request, measure, loopTimeoutDuration);
+ return createReusableExecutionInternal(request, measure, loopTimeoutDuration, hints,
+ extensionNameToPrefix);
#endif
}
@@ -159,13 +169,16 @@
nn::GeneralResult<nn::SharedExecution> ResilientPreparedModel::createReusableExecutionInternal(
const nn::Request& request, nn::MeasureTiming measure,
- const nn::OptionalDuration& loopTimeoutDuration) const {
+ const nn::OptionalDuration& loopTimeoutDuration,
+ const std::vector<nn::TokenValuePair>& hints,
+ const std::vector<nn::ExtensionNameAndPrefix>& extensionNameToPrefix) const {
if (!isValidInternal()) {
return std::make_shared<const InvalidExecution>();
}
- const auto fn = [&request, measure,
- &loopTimeoutDuration](const nn::IPreparedModel& preparedModel) {
- return preparedModel.createReusableExecution(request, measure, loopTimeoutDuration);
+ const auto fn = [&request, measure, &loopTimeoutDuration, &hints,
+ &extensionNameToPrefix](const nn::IPreparedModel& preparedModel) {
+ return preparedModel.createReusableExecution(request, measure, loopTimeoutDuration, hints,
+ extensionNameToPrefix);
};
return protect(*this, fn);
}